Fault tolerant distributed garbage collection system and method for collecting network objects

摘要

A distributed computer system includes a multiplicity of concurrently active processes. Each object is owned by one process. Objects are accessible to processes other than the object's owner. Each process, when it receives a handle to an object owned by any other process, sends a first "dirty" message to the object's owner indicating that the object is in use. When a process permanently ceases use of an object handle, it sends a second "clean" message to the object's owner indicating that the object is no longer in use. Each object's owner receives the first and second messages concerning usage of that object, stores data for keeping track of which other processes have a handle to that object and sends acknowledgement messages in return. The receiver of an object handle does not use the handle until its first message is acknowledged. Periodically, the object's owner sends status request messages to other processes with outstanding handles to that object to determine if any of those processes have terminated and updates its stored object usage data accordingly. A garbage collection process collects objects for which the usage data indicates that no process has a handle. The first and second messages include sequence numbers, wherein the sequence numbers sent by any process change in value monotonically in accordance with when the message is sent. Object owners ignore any message whose sequence number indicates that it was sent earlier than another message for the same object that previously received from the same process.